Abstract:X-ray micro-computed tomography (micro-CT) has been widely leveraged to characterise pore-scale geometry in subsurface porous rock. Recent developments in super resolution (SR) methods using deep learning allow the digital enhancement of low resolution (LR) images over large spatial scales, creating SR images comparable to the high resolution (HR) ground truth. This circumvents traditional resolution and field-of-view trade-offs. An outstanding issue is the use of paired (registered) LR and HR data, which is often required in the training step of such methods but is difficult to obtain. In this work, we rigorously compare two different state-of-the-art SR deep learning techniques, using both paired and unpaired data, with like-for-like ground truth data. The first approach requires paired images to train a convolutional neural network (CNN) while the second approach uses unpaired images to train a generative adversarial network (GAN). The two approaches are compared using a micro-CT carbonate rock sample with complicated micro-porous textures. We implemented various image based and numerical verifications and experimental validation to quantitatively evaluate the physical accuracy and sensitivities of the two methods. Our quantitative results show that unpaired GAN approach can reconstruct super-resolution images as precise as paired CNN method, with comparable training times and dataset requirement. This unlocks new applications for micro-CT image enhancement using unpaired deep learning methods; image registration is no longer needed during the data processing stage. Decoupled images from data storage platforms can be exploited more efficiently to train networks for SR digital rock applications. This opens up a new pathway for various applications of multi-scale flow simulation in heterogeneous porous media.
Abstract:There are inherent field-of-view and resolution trade-offs in X-Ray micro-computed tomography imaging, which limit the characterization, analysis and model development of multi-scale porous systems. In this paper, we overcome these tradeoffs by developing a 3D Enhanced Deep Super Resolution (EDSR) convolutional neural network to create enhanced, high-resolution data over large spatial scales from low-resolution data. Paired high-resolution (HR, 2$\mu$m) and low resolution (LR, 6$\mu$m) image data from a Bentheimer rock sample are used to train the network. Unseen LR and HR data from the training sample, and another sample with a distinct micro-structure, are used to validate the network with various metrics: textual analysis, segmentation behaviour and pore-network model (PNM) multiphase flow simulations. The validated EDSR network is used to generate ~1000 high-resolution REV subvolume images for each full core sample of length 6-7cm (total image sizes are ~6000x6000x32000 voxels). Each subvolume has distinct petrophysical properties predicted from PNMs, which are combined to create a 3D continuum-scale model of each sample. Drainage immiscible flow at low capillary number is simulated across a range of fractional flows and compared directly to experimental pressures and 3D saturations on a 1:1 basis. The EDSR generated model is more accurate than the base LR model at predicting experimental behaviour in the presence of heterogeneities, especially in flow regimes where a wide distribution of pore-sizes are encountered. The models are generally accurate at predicting saturations to within the experimental repeatability and relative permeability across three orders of magnitude. The demonstrated workflow is a fully predictive, without calibration, and opens up the possibility to image, simulate and analyse flow in truly multi-scale heterogeneous systems that are otherwise intractable.